The invention relates to a tobacco shred component recognition method and device based on a convolutional neural network, and belongs to the technical field of tobacco shred component recognition. A set quantity of cut laminas, a set quantity of cut stems, a set quantity of expanded cut laminas and a set quantity of reproduced tobacco cut lamina samples are collected respectively, and all the sample images are cut into images of a set size; the convolutional neural network is established, a certain quantity of cut laminas, a certain quantity of cut stems, a certain quantity of expanded cut laminas and a certain quantity of reproduced tobacco cut lamina samples are selected from the cut sample images as a training set, and the training set is utilized to conduct iterative training on the established convolutional neural network; finally each to-be-recognized tobacco shred image is cut into N images with a set size, and the N cut images are input into the trained convolutional neural network to achieve recognition of the to-be-recognized tobacco shred image. By collecting complete tobacco shred component characteristics, according to the complete tobacco shred component characteristics and through the convolutional neural network, tobacco shred components are recognized, and the accuracy of tobacco shred component recognition is improved.